US 7391357 B1 Abstract Motion measurement errors that extend beyond the range resolution of a synthetic aperture radar (SAR) can be corrected by effectively decreasing the range resolution of the SAR in order to permit measurement of the error. Range profiles can be compared across the slow-time dimension of the input data in order to estimate the error. Once the error has been determined, appropriate frequency and phase correction can be applied to the uncompressed input data, after which range and azimuth compression can be performed to produce a desired SAR image.
Claims(7) 1. A method of producing a SAR image, comprising:
providing input data;
forming a SAR image based on the input data, including range compressing at least some of the input data according to a first range resolution factor;
applying a focusing algorithm to the SAR image to obtain information indicative of a motion measurement error associated with collection of the input data;
using the motion measurement error information to modify the input data; and
forming a further SAR image based on the modified input data, including range compressing the modified input data according to a second range resolution factor that defines a higher range resolution than said first range resolution factor.
2. The method of
3. The method of
4. The method of
5. The method of
6. An apparatus for producing a SAR image, comprising:
an input for receiving input data;
a SAR image former coupled to said input for forming a SAR image based on the input data, said SAR image former range compressing at least some of the input data according to a first range resolution factor;
an error extractor coupled to said SAR image former for applying a focusing algorithm to said SAR image to obtain information indicative of a motion measurement error associated with collection of the input data;
a corrector coupled to said error extractor and said input for modifying the input data based on the motion measurement error information; and
a SAR image former coupled to said corrector for forming a further SAR image based on the modified input data, said last-mentioned SAR image former range compressing the modified input data according to a second range resolution factor that defines a higher range resolution than said first range resolution factor.
7. The apparatus of
Description This application claims the priority under 35 U.S.C. §119(e)(1) of co-pending provisional application Ser. No. 60/665,318 filed Mar. 25, 2005 and incorporated herein by reference. This invention was developed under Contract DE-AC04-94AL8500 between Sandia Corporation and the U.S. Department of Energy. The U.S. Government has certain rights in this invention. The invention relates generally to synthetic aperture radar (SAR) and, more particularly, to correction of motion measurement errors in SAR images. The documents listed below are incorporated herein by reference: - [1] J. L. Walker, “Range-Doppler Imaging of Rotating Objects,” IEEE Trans. on Aerospace and Electronic Systems, AES-16 (1), 23-52, (1980).
- [2] D. E. Wahl, P. H. Eichel, D. C. Ghiglia, C. V. Jakowatz Jr., “Phase Gradient Autofocus—A Robust Tool for High Resolution SAR Phase Correction”, IEEE Transactions on Aerospace and Electronic Systems, Vol. 30, No. 3, pp. 827-834, July, 1994.
- [3] C. V. Jakowatz Jr., D. E. Wahl, P. H. Eichel, D. G. Ghiglia, P. A. Thompson, Spotlight-Mode Synthetic Aperture Radar: A Signal Processing Approach, ISBN 0-7923-96774, Kluwer academic Publishers, 1996.
- [4] W. G. Carrara, R. S. Goodman, R. M. Majewski, Spotlight Synthetic Aperture Radar Signal Processing Algorithms, ISBN 0-89006-728-7, Artech House, Inc., 1995.
- [5] M. Denny, I. Scott, “Anomalous Propagation Limitations to High-Resolution SAR Performance”, Proceedings of the 2002 IEEE Radar Conference, Long Beach, Calif., USA, p. 249-254, 22-25 Apr. 2002.
- [6] Bryan L. Burns, J. Thomas Cordaro, “Imaging synthetic aperture radar”, U.S. Pat. No. 5,608,404, Mar. 4, 1997.
Synthetic Aperture Radar (SAR) forms images of a scene by sampling energy from a scattered field along the radar's flight path and coherently processing the data. Coherence of the data set is facilitated by very accurately measuring the geometric relationship between the desired target scene and the radar's flight path, and accounting for this in the data processing. This requires measuring the radar's motion, or at least its relative motion, very accurately and with fractional-wavelength precision over the course of the synthetic aperture. Typically, an Inertial Measurement instrument is employed, and even this is often aided by Global Positioning Satellite navigation readings. The raw SAR data is typically a two-dimensional array of complex data samples, with one dimension representing samples from echoes of individual pulses (fast-time), and the other dimension representing the pulse index number (slow-time). This collection is termed the phase history data. Since wideband modulation techniques, such as the Linear Frequency Modulated (LFM) chirp waveform, are normally used for individual pulses, the data needs to be processed, or compressed, in the intra-pulse or range direction to achieve the final desired range resolution. This is termed range-compression. The data needs further processing in the inter-pulse or azimuth direction to complete the image formation process. This is termed azimuth compression. During the course of a synthetic aperture, as the radar's perspective towards a target scene changes, ranges to some target locations change or migrate relative to other target locations. This migration is deterministic and is compensated within the image formation process by algorithms such as the Polar Format Algorithm (PFA) developed by Walker (document [1] above). Relatively small motion measurement errors manifest themselves principally as phase errors in the complex data samples, and if large enough become observable as a smearing, blurring, or other degradation in the image. For most SAR systems, however, the nature and degree of blurring is nearly identical in different parts of the degraded SAR image. This allows a measurement of the blurring function, and then a calculation of a suitable correction to be applied to the original data to compensate for the presumed motion error. Further processing then may yield a well-focused image devoid of the previously observable degradation. A number of algorithms exist to automatically focus the degraded image. While some measure and compensate blurring, others seek to optimize other measures, such as contrast ratio in the image. Collectively, these processes are termed “autofocus” algorithms. A very popular autofocus algorithm is the Phase Gradient Autofocus (PGA) algorithm described by Wahl, et. al. (document [2] above). Very large relative motion measurement errors manifest themselves as an unexpected additional shifting or migration of target locations beyond the aforementioned deterministic migration during the course of the synthetic aperture. Degradation in images from data exhibiting errors of this magnitude are substantial, often rendering the image useless. Application of conventional autofocus techniques are unable to properly mitigate the image degradation. A general presumption in the SAR community is that any motion measurement errors are less than the range resolution of the radar. Under this presumption, autofocus operations are conventionally applied to fully range-compressed images. Since autofocus typically requires iteratively processing the data into an image, efficiency is gained by repeating only the azimuth compression, and not the range compression operations. This presupposes that, for example, a radar with 2 cm nominal wavelength and 30 cm range resolution will never see more than (4π/λ) ρ In addition to motion measurement errors, longer ranges impart greater deleterious atmospheric effects to the data, whereby electrical path lengths depart significantly from the physical path lengths. The electrical path length is related to the actual path length by the ratio of the average wavelength to the nominal wavelength, and accounts for atmospheric dielectric variations, refraction and other wave propagation phenomena. Since coherence depends on electrical path lengths, problematic errors similar to motion measurement errors may be induced by perturbations in the atmosphere's transmission characteristics in spite of perhaps otherwise adequate motion measurements. While the presumption that apparent range errors are less than the radar's range resolution is often true, modern high-performance SARs can exceed this criterion. The drive for finer resolutions, longer ranges, and less expensive (and less accurate) motion measurement systems will increasingly cause situations where a target's echo return effectively exhibits a residual migration error exceeding one or more range resolution cells during the course of the synthetic aperture. This would doom to failure any autofocus scheme that presupposes otherwise, which includes autofocus schemes that operate only on fully range-compressed data. An example of this situation is illustrated by It is desirable in view of the foregoing to provide for correction of motion measurement errors that extend beyond the range resolution of a SAR. Exemplary embodiments of the invention can achieve this by effectively decreasing the range resolution of the SAR in order to permit measurement of the error. Other exemplary embodiments of the invention compare range profiles across the slow-time dimension in order to estimate the error. Once the error has been determined, appropriate frequency and phase correction can be applied to the uncompressed input data, after which range and azimuth compression can be performed to produce a desired SAR image. The present invention recognizes that a phase error function in the azimuth direction cannot be ascertained from a fully range-compressed data set, since the error energy is spread across several range resolution widths. Therefore, some embodiments impose the constraint that, for extracting an autofocus correction vector, the range resolution must be coarse enough to encompass the phase error. Put another way, the phase error is measured on data that is not fully range-compressed, i.e. radar data with degraded range resolution. This can be done in some embodiments by using only part of each return echo, that is, a portion of the fast-time vector. It can also be done in some embodiments by blurring fully range-compressed data in the range dimension. In embodiments that use range subapertures for image formation, a single range subaperture can be employed for phase-error measurement. Once an accurate phase error has been measured, then the corresponding migration effects can be calculated. Some embodiments determine the actual migration effects by correlating range-compressed pulses with each other. This process of correlating range profiles obviates the need for identifying and selecting a prominent scatterer, allowing improved performance on SAR images that lack prominent scatterers. Once the migration effects have been adequately characterized, compensation can then be applied to the SAR data. The excessive range migration should be mitigated, that is, excessive range shifts in range-compressed data should be eliminated. The echo returns should be shifted back into proper position. Range shifts in range-compressed data are achieved in some embodiments by multiplying the uncompressed data with a fast-time-dependent phase shift, that is, a complex sinusoid that shifts frequency in addition to phase. Assume that the target is an isotropic point scatterer, and the waveform is a Linear Frequency Modulated (LFM) chirp. The radar echo from a point target may be adequately described in the phase history data by A T=pulse width of the radar, n=the slow-time pulse index number with N/2≦n<N/2, ω γ t t=time, c=the velocity of propagation, r r The radar's motion measurement system measures r
Particularly problematic is the line-of-sight component of the motion measurement error, that we denote
Other error components have minor impact on SAR image quality by comparison.
After sampling, the digital data set is described by By making the usual assumption of a flat image plane, we identify s s ψ ψ α By further noting that the first occurrence of ε dα=a nominal azimuth sampling frequency scale factor, ω γ T i′=the new fast-time sampling index with I′/2≦i′<I′/2, n′=the new slow-time sampling index with N′/2≦n′<N′/2. While this corrects for deterministic migration, it also impacts the nature of ε
For typical resolutions where the range of an is small, this approximates to
A byproduct of the sampling/resampling is typically that ω
Data corrected for deterministic migration to facilitate image formation using for example the Polar Format Algorithm yields
As a practical matter, it is often adequate to assume,
By recognizing that the achievable nominal resolutions of this data set are A first order expansion of ε′
This allows the data model to be rewritten as
Recall that the Discrete Fourier Transform (DFT) of a finite-length complex exponential is given by Range compression on our data model entails performing a DFT over index i′. This yields
Although this expression was developed by range-compressing the resampled phase history data, it should also be obvious that the same form for the data can be achieved by beginning with an image and undoing the azimuth compression step. Furthermore, resampled pseudo-phase history data can be generated by undoing the range compression as well. This expression has a peak response in range at the index value
This is the customary presumption for autofocus algorithms, as it allows the further approximation Exemplary embodiments of the invention can mitigate the effects of ε Exemplary embodiments of the invention measure a phase error, and calculate a corresponding range-shift. Then phase and frequency corrections are applied to the data to correct both. If equation (26) is satisfied, then ε If equation (26) is violated, various embodiments process the data to a new coarser range resolution p′ The coarser range-resolution may be accomplished in any suitable manner. For example, the complex data can be filtered or blurred in the range dimension, as shown generally at Once the motion error ε
Data corrected this way may be processed into an image in the usual manner, for example with a two-dimensional Fourier transform. Note that the data correction is both a fast-time frequency shift and a phase shift. While typical autofocus algorithms are iterative for optimum performance, the frequency correction of a single iteration is adequate in some embodiments, although other embodiments also use iterative phase correction to derive additional benefit from further iterations. Once residual migration effects are contained within a range resolution cell, then some embodiments use conventional iterative autofocus techniques to “finish” the job. In some embodiments, the process begins with the phase history data as shown at Although the embodiments described above base their corrections on the determination of the quantity ε Some embodiments measure a range shift directly in the range-compressed data, and then compensate the input data with phase and frequency corrections. Tracking a point-like target in the range-compressed data (as in the prior art “Prominent Point Processing” of document [4] above) is inadequate in many instances because prominent point-like targets may not exist in the field of view of the radar. Exemplary embodiments of the invention correlate the entire range profile to establish a shift-gradient in the slow-time dimension. Note that the range profile of a single point scatterer is given by
As the error varies on a pulse-to-pulse basis, so does the peak value position for index v. Define the total apparent shift as Comparing the profiles for different pulses n′, e.g., profiles In practice, the pulse-to-pulse error gradient Δε Because the gradient Δε
Some embodiments select n Once a gradient has been ascertained, an accumulation of the gradients yields an approximation of the actual function ε The gradient processing described above is illustrated generally at The frequency correction to the pre-range-compressed data, shown at Referring to the example of In some situations, phase history data may not be available or convenient. Undoing the azimuth and range compression operations on a complex image (as described above) provides an equivalent to resampled phase history data that will generally suffice for this purpose. In this manner, some embodiments permit even an image that has had conventional autofocus algorithms applied to it to be further corrected by the techniques of Excessive residual migration due to motion errors, or apparent motion errors due in fact to atmospheric propagation phenomena, are not always correctable with conventional autofocus algorithms. Exemplary embodiments of the invention apply corrections before final range compression. Some embodiments apply both a frequency correction as well as a phase correction before final range compression. Some embodiments correct excessive migration in polar-reformatted data by correcting for both the motion error and the azimuthal derivative of the motion error. Some embodiments apply only a properly ascertained phase correction, which can improve the image. Some embodiments measure excessive migration by performing conventional autofocus steps on a reduced-range-resolution image, provided the reduced range-resolution is coarser than the residual migration. Excessive migration can also be estimated in some embodiments by correlating range profiles in range-compressed data. Sensitivity to range profile shifts, and hence migration, can be enhanced in some embodiments by oversampling the range compressed data in the range dimension, and in other embodiments by correlating range profiles that are separated in the slow-time dimension. Workers in the art will recognize that the exemplary embodiments described above can be readily implemented by, for example, suitable modifications in the hardware, software, or both hardware and software, of conventional SAR image production systems, such as DSP-based systems. Although exemplary embodiments of the invention have been described above in detail, this does not limit the scope of the invention, which can be practiced in a variety of embodiments. Patent Citations
Non-Patent Citations
Referenced by
Classifications
Legal Events
Rotate |